IMECE DEVELOPING A BASELINE STRATEGY FOR CONTROLLING BLINDS IN BUILDINGS

Transcription

1 Proceedings of the ASME 2013 International Mechanical Engineering Congress & Exposition IMECE 2013 November 15-21, 2013, San Diego, USA IMECE DEVELOPING A BASELINE STRATEGY FOR CONTROLLING BLINDS IN BUILDINGS Nasim Karizi T. Agami Reddy The Design School Arizona State University Tempe, Arizona nkarizi@asu.edu Partha Dasgupta School of Computing Informatics Arizona State University Tempe, Arizona ABSTRACT Automated day-lighting systems can modulate window blinds and electrical lights for maintaining the proper illumination levels and save significant electrical energy in buildings. This paper presents an initial work toward developing an automated installation and maintenance of a generic daylighting system which is able to self-calibrate and adapt to the building needs with minimal human intervention. The system operates based on information provided by a wireless sensor network, and processed through learning algorithms and feedback control principle. This paper focuses on a preliminary simulation study to establish a control baseline and identifies the required elements. It demonstrates the concept, using daylighting simulation software in the context of a test cell which represents a virtual office space. A startup baseline for the optimal blind slat angle settings for the windows is developed with the objective of maintaining uniform lighting levels on a horizontal surface inside the test cell. The lighting baseline simulations are limited to specific times and days of a year to reduce and optimize the simulation process and are applied to predict the optimal blind slat angles for other days of the year. This paper presents and discusses the results of such an analysis including an extrapolation to all year round. Address all correspondence to this author. INTRODUCTION Every facet of human society is affected by the operational costs of buildings. Offices, businesses, data centers, homes and so on consume power and manual attention. There are 1.5 million commercial buildings (and 114 million homes) in the US with the growth rate at about 3% a year. Energy savings are of primary interest for societal, but lowering the cost of deployment and maintenance of smart control systems is a close second, [1 4]. Furthermore, adaptive automated daylighting control systems have immense potential to significantly reduce energy consumption and avoid the glare issues inside buildings, especially high performance buildings. In addition to the use of high levels of insulation, energy efficient windows, low levels of air infiltration and heat recovery ventilation; the design of such buildings involve a number of advanced elements and sub-systems that require dynamic control during operation (such as advanced fenestration systems, daylight harvesting, demand ventilation control, advanced HVAC system control, heat recovery systems, groundsource heat pumps and so on). Various decision-making capabilities at the Energy Management and Control System level (EMCS) are being developed for these buildings such as fault detection, supervisory control, load shedding in response to realtime electricity pricing; which greatly enhance the value of such buildings, [5]. However, these current methods of control use fixed programs and are expensive to deploy, maintain, re-program and do not dynamically change to varying physical conditions. In addition, they need enhancement to improve reli- 1 Copyright c 2013 by ASME

2 ability and poor performance in some cases, [6]. Due to the complex interaction between artificial lighting, daylighting, solar heat gain through windows and thermal comfort based on human factors in the buildings, an accurate estimation of the energy saving values by lighting controllers and their impact on the total energy consumption in a building is very challenging. Several research studies (measurement studies [4, 7, 8] and simulation analyses [9 14]) have shown that by applying proper daylight controllers, up to 60% of lighting energy could be saved. However, these studies have not been able to encourage designers and engineers to apply these kinds of daylighting controllers in their projects, [15]. One of the main reasons is the poor performance of the current controllers which leads to lower energy saving values compared to predicted savings conducted by energy simulation models. In addition, most of the current energy simulation tools are not able to properly model the actual performance of daylighting systems in indoor environment of the buildings [9 14]. This fact also leads to unreliable energy saving results by simulation software. OBJECTIVE AND SCOPE Our main goal of conducting this research study is to develop an adaptive control method to optimally operate buildings of every size and vintage. By introducing self-managing algorithms into the effort, we seek to enhance the capabilities of control systems by almost completely automating their deployment and operation and by empowering them to adapt to environmental changes. During its operation, the system will track occupancy, manual overrides and, utility company s demandresponse guidelines. Learning algorithms are then used to further enhance the functionality of the system. This paper demonstrates the basic design step for developing such a control system by demonstrating a simulation based approach for daylighting control inside a virtual office space and is meant to serve as a baseline for the learning algorithm. In the further steps, the developed learning algorithms will be applied on an actual test environment to evaluate and refine their performance. METHODOLOGY Daylighting Simulation The test room model was built in ArchiCAD and imported to Relux Professional lighting software, [16] which uses average indirect fraction methodology for calculation, [17]. The program provides illuminance levels outputs in excel format which was imported to MATLAB for further analyses. Furthermore, a small- scaled virtual test bed model is built to investigate daylighting, glare issues and required artificial lighting in a virtual office environment. The model pertains to a feet office area with three 1 by 1.3 inches windows on the North, West and South walls as shown in Fig. 1. All interior walls and ceiling are from Birch wood and the interior frames of the windows are made of matte aluminum. All three windows are equipped with matte aluminum blinds which can control the amount of daylight penetrating to the test cell through different slat angle settings. Furthermore, to calculate and demonstrate the light level on the small scaled work area, a virtual measurement surface inside the test cell is defined which represents the working surface in the office, (Fig. 1). The calculations were done for Phoenix, Arizona ( N, W) for clear sky according to International Commission on Illumination, (CIE), (less than 30 % cloud cover, or none). As stated earlier, the intent of the baseline simulations is to develop an initial blind control strategy which will be refined by the adaptive algorithm over time. So as to reduce set up times, the simulations are kept to a minimum number of runs. Hence, we perform simulations over a few days over a year which capture the variability of the solar movement over the year and day. Based on this fact, three discrete days and times of year are chosen as following: June 21 : 10 AM, 12 PM and 2 PM September 23 : 10 AM, 12 PM and 2 PM December 23 : 10 AM, 12 PM and 2 PM The daylight simulations were performed including only one open window at a time, (with different blind slat angles) with other two windows totally closed. We assume that illuminance level at a specific area is additive and the total lighting for the case whereas all three windows are open was calculated as the sum of the solar radiation contribution from separate simulation cases including window north, west or south open with different slat angles. All three windows were equipped with interior blinds (matte aluminum) which were movable from 0 degree (fully open) to 90 degree (fully closed) in 15 degree intervals. Analysis Calculation of CV for all blinds setting combinations The results from daylighting simulation models were imported into MATLAB code for analyzing all the possible blind setting combinations. The control objective was to get the illumination levels at all 9 measurement points as uniform as possible, (a target of 250 Lux was assured). To achieve this goal, the Coefficient of Variation, (CV) values for each measurement point for each time/blind setting were calculated. Sum of Square Error: SSE = 9 ( j=1) (y j y i ) 2 Root Mean Square Error: RMSE = CV Value Calculation: CV = RSME ȳ SSE 9 Whereas: y i : desired light level in LUX on the defined work plane inside 2 Copyright c 2013 by ASME

3 FIGURE 1. The virtual test cell environment plan and sensor placement, (top) and Relux model, (bottom) the test room y j : measured light level in LUX by 9 sensors installed on the work plane The histograms in Fig. 2, Fig. 3 and Fig. 4 illustrate the frequency of the CV Values calculated for all blind slat settings combinations, (3 windows and 7 blind stat settings for each window, (0, 15, 30, 45, 60, 75, 90), 73 = 343 combinations) in June 21st, September 23th and December 23th. The graphs indicate that most of the settings have a CV value between 0.7 and However, in each case, there is one best CV value which is close to 0 and demonstrates the best unified illuminance level on the work plane. FIGURE 2. Histogram graphs for CV values for all possible blind setting combinations. Table 1 displays the best combinations for all three specific simulation days, (June 21st, September 23th and December 23th) based on smallest CV values including the best blind angle setting for all three windows located on North, West and South walls. The average Lux level among the 9 measurement points on the defined work plane in the test cell is also demonstrated in the last column of the table. 3 c 2013 by ASME Copyright

4 FIGURE 3. Histogram graphs for CV values for all possible blind setting combinations. FIGURE 4. Histogram graphs for CV values for all possible blind setting combinations. Cost function calculation for all combinations One of the important aspects in controlling the blinds in an optimized setting is visual comfort. The amount of change in the blind slat angle in each setting needs to be minimized to avoid the distraction caused by blinds motor voice and visual discomfort. To achieve this goal, cost function was developed that in- cludes a weight factor α to define the optimal settings based on the best CV values and least slat angle changes of the blinds for the next setting. 4 c 2013 by ASME Copyright

5 June W(N) W(W) W(S) CVs Lux Ave. 10:00 AM :00 PM :00 PM September 10:00 AM :00 PM :00 PM December 10:00 AM :00 PM :00 PM TABLE 1: The bhtst CV Values and their blind s angle settings for June, September and December. This function is defined as below: ( ((Ai A x ) 2 +(B i B x ) 2 +(C i C x ) 2 ) ) Cost=(1 α)cv+α ( ) (1) Where the variables are defined as follows: α: weight factor A i : current blinds slat angle for window A (north-side) B i : current Blinds slat angle for window B (west-side) C i : current blinds slat angle for window C (south-side) A x : optimal blinds slat angle for window A B x : optimal blinds slat angle for window B C x : optimal blinds slat angle for window C In this equation α is defined as a weight factor between CV values and slat angle changes which has to be chosen properly. When weight factor α = 0 the cost function value equals to the best CV value without taking the slat angle changes into consideration. For α = 1, the CV value weight factor is ignored and the objective would be to maintain the blind slat angle settings constant and close to keep the cost function value close to zero. since the amount of changes in blind slat angles are not significantly important for yearly prediction. Based on this fact, the cost function will be just based on the best CV values which are calculated where the initial condition of all blinds is set to zero [A=0,B=0, C=0]. The CV values show that the simulated months, (June, September and December) could be used to predict the best blind settings for at least two months before and two months after at the same time. The graphs in Fig. 5 illustrate the CV values, when the best setting in June 21st, 10 AM, 12 PM and 2 PM are applied to the same time of other months. In June, at 10 AM the best settings for the blinds based on best CV values are [0, 90, 0]. The same setting has been applied on all other 11 month to find out how well the best setting in June works out as an optimal set up for the rest of the year. The results from daylighting simulation confirms that the CV values for April, May, July and August are very close to the best CV value in June. This fact shows that the best blind settings for June 10 AM, [0 90 0] could be applied as one of the best setting s options on May, April, July and August 10 AM, since the difference between CV values is not significant. The same methodology applies for 12pm and 2pm each year. The graphs in Fig. 6 illustrate the yearly prediction based on December 23rd, at 10AM, 12PM and 2PM. As the graphs show, similar to June, the best angle settings in December could be used to predict the best blind settings for at least two month before and two month after December. For example, the best simulated setting for December 10AM is [45, 90, 0]. The same setting is applied to all other 11 months and the results in form of CV values are shown in the graph which indicates that this setting is one of the best setting s options on May, April, July and August at 10 AM, since the difference between CV values is not significant. The same methodology applies for 12PM and 2PM. As noted in Fig. 5 and Fig. 6, the difference between CV values for predicted settings based on the best simulated setting in June and also December is no greater than 2%. Based on this fact, to have more accurate yearly prediction results, the simulated months need to be extended to three months a year. OPTIMAL CONTROL SETTINGS PREDICTION As explained before, the number of simulations is limited to June, September and December, 10 AM, 12 PM and 2 PM. The following section of the paper explains how to predict the best slat angles settings of the blinds for the three windows; (North (A), West (B) and South (C)) based on cost function values. The prediction phase is divided into two different sections: Yearly prediction and Daily prediction. Yearly Prediction of the Best Blind s Slat Angles For yearly predictions of the best blind settings, the weight factor α in the cost function equation is equal to zero, (α = 0) Daily prediction of the best blind s slat angles For daily predictions, choosing the weight factor α in the cost function is very important. Based on this criteria, the impacts of different values of α on the cost function has been investigated and the best blind settings, (for three A, B and C windows) resulted from the smallest cost function values for June and December have been compared and shown in Fig. 7 and Fig. 8. As illustrated in Fig. 7 and Fig. 8, the graphs indicate that for both months, α = 0.25 is the best value which provides very close blind slat settings in comparison to the best settings with α = 0 which is based on best CV values for each case. Therefore, 5 Copyright c 2013 by ASME

6 FIGURE 5. Four months prediction method based on June 21st, at 10 AM, 12 PM and 2 PM. FIGURE 6. Four months prediction method based on December 23th, at 10 AM, 12 PM and 2 PM. the weight factor is chosen equal to 0.25 in the cost function, ( ((Ai A x ) 2 +(B i B x ) 2 +(C i C x ) 2 ) ) Cost = 0.75CV ( ) (2) The following graphs demonstrate the daily prediction method for the whole day from 9 AM to 4 PM based on the minimized cost function values from equation (2). As explained earlier, one of the main goals of the research study is to predict the best blind s angle settings based on limited simulations which in this case are limited to two days in the year, (June 21 and December 23), and three times of the day, (10 AM, 12 PM and 2 PM). Graphs in Fig. 9 and Fig. 10 show the predicted values for the whole day based on 3 hours simulated values. For all windows, the predicted angle settings line is following the simulated setting lines symmetrically. To validate the predicted angles, the exact same blind setting read from the predicted line in the graphs for specific times are simulated and their CV values are compared to the best CV values of those specific times. Tab. 2 assembles the analysis results. 6 Copyright c 2013 by ASME

7 FIGURE 7. Comparison between the best blind s settings based on different values of α for June 21, (10 AM, 12 PM and 2 PM). FIGURE 8. Comparison between the best blind s settings based on different values of α for December 23, (10 AM, 12 PM and 2 PM). Time CV W(A) W(B) W(C) Simulated 9:00AM Predicted 9:00AM Simulated 11:00AM Predicted 11:00AM Simulated 13:00AM Predicted 13:00AM Simulated 11:00AM Predicted 11:00AM Simulated 13:00AM Predicted 13:00AM TABLE 2: Validation Table assembling the predicted CV values with the best CV values resulted from simulations for June with α = Time CV W(A) W(B) W(C) Simulated 9:00AM Predicted 9:00AM Simulated 11:00AM Predicted 11:00AM Simulated 13:00AM Predicted 13:00AM Simulated 11:00AM Predicted 11:00AM Simulated 13:00AM Predicted 13:00AM TABLE 3: Validation Table assembling the predicted CV values with the best CV values resulted from simulations for December α = Copyright c 2013 by ASME

8 FIGURE 9. Predicted blind s settings based on best settings for α= 0.25, for June 21, (10 AM, 12 PM and 2 PM). FIGURE 10. Predicted blind s settings based on best settings for α= 0.25, for December 23, (10 AM, 12 PM and 2 PM). As shown in Table 2 and Table 3, for June 21 and December 23, the CV values of predicted settings for 9 AM, 11AM, 1PM, 3PM and 4PM are compared with the best CV values resulted directly from simulation on the same days and hours. The compared values indicate a very insignificant difference between best simulated CV values and the predicted values, ( 0.08). CONCLUSIONS AND FUTURE WORK The analysis results indicate that the predicted blind settings based on daily and yearly prediction methodologies are acceptable as initial baseline settings inside the test environment which then will be refined and optimized by the adaptive control algorithm. This baseline will spare us a significant amount of time in fine-tuning the control algorithm in the experimental steps. 8 Copyright c 2013 by ASME

9 As explained earlier, our main goal of conducting this research study is to design a self-managing algorithm to enhance the capabilities of control systems by almost completely automating their deployment and operation by empowering them to adapt to environmental changes. Based on this fact, future work would include investigation of impacts of artificial lights inside the virtual test cell. Furthermore, as a part of our design approach we will implement and validate the developed algorithm on an actual test-cell (of size ft) which is equipped with controllable smart blinds. In the last design step, the system will be implemented on a real office building for the final refinements. REFERENCES [1] Lee, E. S., Bartolomeo, D. L. D., and Selkowitz, S. E., The effect of venetian blinds on daylight photo electric control performance. Journal of the Illuminating Engineering Society, 28, pp [2] Rubinstein, F., Jenning, J., Avery, D., and Blanc, S., Preliminary results from an advanced lighting controls test bed. Journal of the Illuminating Engineering Society, 28, pp [3] Choi, A., and Sung, M., Development of a daylight responsive dimming system and preliminary evaluation of system performance. Building and Environment, 35, pp [4] Lee, E. S., and Selkowitz, S. E., The new york times head quarters daylighting mockup: monitored performance of the daylighting control system. Energy and Buildings, 38, pp [5] Beldi, S., and Krarti, M., Genetic algorithm based daylight controller. In Proceedings of the Inaugural US- EU-China. Thermophysics Conference UECTC-RE. [6] Seoa, D., Ihm, P., and Krarti, M., Development of an optimal daylighting controller. Building and Environment, 46, pp [7] Li, D., Lam, T., and Wong, S. L., Lighting and energy performance for an office using high frequency dimming controls. Energy Conversion and Management, 47(9 10), pp [8] Atif, M. R., and Galasiu, A. D., Energy performance of daylight-linked automatic lighting control systems in large atrium spaces: report on two field monitored case studies. Energy and Buildings, 35(5), pp [9] Bjorn, B., and Eilif, H. H., Energy savings in lighting installations by utilizing daylight. CADDET energy efficiency newsletter, 1. [10] Jonathan, M., Patrick, J. B., and Doug, C. H., Ashrae journal. Energy and Buildings, 40(5), pp [11] D. Li, J. C. L., Evaluation of lighting performance in office buildings with daylighting controls. Energy and Buildings, 33, pp [12] Krarti, M., Erickson, P. M., and Hillman, T. C., A simplified method to estimate energy savings of artificial lighting use from daylighting. Building and Environment, 40(6), pp [13] Chel, A., Tiwari, G. N., and Chandra, A., A model for estimation of daylight factor for skylight: an experimental validation using pyramid shape skylight over vault roof mud house in new delhi india. Applied Energy, 86(11), pp [14] Chel, A., Tiwari, G. N., and Singh, N. H., A modified model for estimation of daylight factor for skylight integrated with dome roof structure of mud-house in new delhi, india. Applied Energy, 87(10), pp [15] Krarti, M., Energy audit of building systems: an engineering approach. CRC Press and Francis and Taylor Group, Boston, MA. [16] Relux Informatik AG, Relux light simulation tools, January. [17] Shikder, S. H., Price, A., and Mourshed, M., Evaluation of four artificial lighting simulation tools with virtual building reference. In European Simulation and Modelling Conference. 9 Copyright c 2013 by ASME